Skip to content

Distribute and run LLMs with a single file.

Commits on main

Clarify 'man' as manual in README (#376)
Matt committed 2 years ago
564d9fb
Sync with upstream llama.cpp project
Justine Tunney committed 2 years ago
94d0940
Fix issues with server send_embeddings function
Justine Tunney committed 2 years ago
0e2845a
Fix memory bug in grammar parser
Justine Tunney committed 2 years ago
22aba95
Upgrade to latest cosmocc
Justine Tunney committed 2 years ago
68aef2b
Fix server multimodal statistics (#392)
CJ Pais committed 2 years ago
a2d159e
Revert moondream vision language model support
Justine Tunney committed 2 years ago
aa8c01a
Create tool for synchronization reports
Justine Tunney committed 2 years ago
8dae978
Change the way the new fast/precise flags work
Justine Tunney committed 2 years ago
b749326
Fix vim modelines (#351)
Jōshin committed 2 years ago
911d58f
eecbf89
Add special tokens to server llama_decode() inputs
Justine Tunney committed 2 years ago
7900294
Add CLI tool for printing embeddings
Justine Tunney committed 2 years ago
42bd9b8
Upgrade to cosmocc v3.3.4
Justine Tunney committed 2 years ago
6c45e3e
Add --precise and --fast flags
Justine Tunney committed 2 years ago
bbae0f6
Speed up prediction on CPUs with many cores
Justine Tunney committed 2 years ago
89c189e
Add vectorized expf() functions
Justine Tunney committed 2 years ago
e2b3cb2
Disable an unintended integrity check
Justine Tunney committed 2 years ago
2b4da98
Make GGML vector ops go faster across hardware
Justine Tunney committed 2 years ago
c9d7393
CUDA: CUDART < 11.7 workaround for __hmax, __hmax2
Johannes Gäßler committed 2 years ago
38311f2
Shave 9 seconds latency off tinyBLAS compilation
Justine Tunney committed 2 years ago
2313ac7
Upgrade to latest llama.cpp code
Justine Tunney committed 2 years ago
0bdea60
Use more Kahan summation
Justine Tunney committed 2 years ago
9540b43
Add Kahan summation to tinyBLAS Q8/Q4 on ARM
Justine Tunney committed 2 years ago
2af3b88
Roll back GGML barrier refactorings
Justine Tunney committed 2 years ago
540dc16
Synchronize after every op
Justine Tunney committed 2 years ago
9cf7363
Use 4.7x fewer synchronization barriers in GGML
Justine Tunney committed 2 years ago
bd8c0de
Rewrite synchronization primitives
Justine Tunney committed 2 years ago
6162004